Mixture Density Networks-Based Knock Simulator

This paper proposes a statistical simulator for the engine knock based on Mixture Density Network (MDN) and accept-reject method. The proposed can generate random intensity signal corresponding to input signal. generated has consistent probability distribution with real engine. Firstly, analysis is conducted experimental data. From results, some important assumptions properties of are made. Regarding as variable discrete-time index, it independent identically distributed if identical. under identical be approximated by Gaussian Model(GMM). parameter GMM function input. Based these assumptions, two sub-problems establishing formulated: One approximate from parameters an absolutely continuous function; other one design number generator that outputs data given distribution. MDN applied density algorithm used design. method evaluated in data-based validation.